So, can't you add to your plug in a noise pattern remover?

What Gain values are you using??

I don't remember what the gain was on the night shots -- I think there was a little.

The reason I don't add a noise pattern removal system to my plug-in is that there's already one built-in to the camera software. However, it can introduce artifacts too. I'll play with it.

I don't remember well who, but I think I read somewhere here someone was using a gain of 5.0 ....

If you could get a correct balance among Gain, Gamma and contrast, may be you would get better images, but I don't know...

Okay, the link is fixed:

http://home.mindspring.com/~jrod/head_shoulders_B.jpg

The image looks a little softer because I applied Ben's plugin (again), and also added Grain removal in Combustion.

Color-Correction was done with Color Finesse.

So Ben, you're saying that the banding I'm seeing won't move at all? It's a fixed band? It seems as though there would be some noise there.

I'm curious to see what would happen if you were back-lit, so that your face wouldn't be the hottest part of the scene, and you didn't clip the rim on the backside of your head.

Still has a green cast.

To everybody,
Don't be too afraid of color noise, it is easily removed without affecting overall image quality.
What is really important is to keep low noise on Luma..(or Green in this case)

Okay, I removed more of the green. Use the same link.

So Ben, That's suppose to be bright back there?

The datasheet on the FillFactory site says that there should be a dynamic range of 64db, that's around 10 f-stops.

"The datasheet on the FillFactory site says that there should be a dynamic range of 64db, that's around 10 f-stops."

We'll see. I'm definitely still getting the hang of it, and who knows if Sumix is taking full advantage of the sensor -- either with the camera or with the software.

I'm doing some tests to increase saturation, and they're promising. I'll post a couple movies in a bit.

- ben

BTW,

If you shrink the image down to NTSC, it looks pretty good, like a frame from a film (although again I have no idea what the contrast ratio is in that scene, film has a lot of dynamic range). I've noticed there's a lot of aliasing, etc. in the full-res version, so I guess there needs to be some anti-aliasing added, or something. Like the circle on the shirt, the edge of your face (light side), etc. A lot of aliasing.

But of course you can't see that on NTSC :-)

http://home.mindspring.com/~jrod/hea...lders_NTSC.jpg

You know, maybe at $1000 for the camera we should consider these "super NTSC" cameras instead of HD cameras. I'm just saying I'm not sure we can get away with all that aliasing and bayer artifacting on an HD monitor.

edit: I got rid of the aliasing by using only the "Y" offset, so the image looks pretty good now. There is some JPEG artifacting though on the red of the shirt, that is NOT aliasing.

No way am I going to this much trouble for NTSC.

I'm figuring out ways to deal with the saturation -- check this out.

There are major artifacts at 100%, but this lets you see the original color and the enhanced color.

Really nice.
What about the Sky, what time was it??
Gain, gamma, contrast?

Could you post a full resolution .jpg frame?

There is also a problem with color balance.Look at the white signal near the tree....still a lot of green all over..
You need a higher gain on blue also...

What are the artifacts? Can they be removed?

Also is there a way you can expose so that the sky doesn't clip, but still get the information on the street? Outdoor scenes with a hot sky (sun right behind the clouds, thin cloud-cover, etc.) are often 9-10 stops, so if you can get that, then you're fine.

I'm working on how to remove the artifacts, or do a more suitable saturation operation.

It was very difficult to expose. I think I might go out and try another. It seemed impossible not to clip the sky, but I'll give it another shot.

The color balance is still a little off, but you get the point.

I posted a movie of the shot here. It's small, but it shows you that at 23-24fps, you can get a very stable, smooth image with no strobing, no rolling shutter artifacts, etc. The exposure time was around 25ms (a little more than 1/48th second). You can crank the exposure time up higher, until you're shooting with the shutter continuously open. Then you start to see rolling shutter take effect.

- ben

Ben, could you give me a little more detailled info?
like gain setting for every color, gamma, contrast, etc.
Also at what time of the day this clip was captured.
Also if it was before 5 p.m most of the time you'll get a white sky, even on film, also it depends on if you have the sun in front of camera or behind...

one of the problems is that all the 1 cmos or 1 ccd cams look like machine vision..not film...they don't have enough color! I hope the Altasense has good color..I know we can get it because my 10D is single cmos and it look GREAT

but maybe we need a chip that was designed for production not machine vision ;) I think the Altasense will do the trick

from what I have seen sofar I think the Micron looks better then that IBI5 you jut got...but then you need to shoot a shot with a blown out spot like a light bulb and see if it streaks the image or not?

is yours 2/3 inch?

But if you go YUV and apply the correct compensation on color you get really good results...


Obin, could you someday post a couple of image sequences (16 bit per color channel or what the camera gives) in Tiff or TGA .
Just to see what are all those problems you talk about, cause I can't get a decent idea of them...

We are having major recording issues with the CIneLink. It's not fast like it should be for record...will keep everyone posted on this as it gets resolved...

Hey Ben,

That movie looks good!

Are the artifacts color fringing, especially on the trees?

Were the same artifacts on the "non" saturated version?

BTW, if that's a non-color-corrected image, then I think you can definitely underexpose the image to get the right exposure for the sky and do CC afterwards to bring up the dark cars, etc.

Sure I will ...I guess I have sent images to Steve but not you guys...I will try and dig that email up and send it your way


<<<-- Originally posted by Juan M. M. Fiebelkorn : But if you go YUV and apply the correct compensation on color you get really good results...


Obin, could you someday post a couple of image sequences (16 bit per color channel or what the camera gives) in Tiff or TGA .
Just to see what are all those problems you talk about, cause I can't get a decent idea of them... -->>>

BTW Obin,

Raw film doesn't look "good" either. If you're going to pack an image with a lot of dynamic range, you're not going to get saturation, and vice versa. Good saturation comes with increased contrast.

Take these images for instance:

http://www.24p.com/asc_web/35mm_girl.jpg

This is a 35mm frame with around 12-13 stops of dynamic range captured. You're not going to get "pretty" images when you do that. Her flesh tones are practically non-existant. Now increase the contrast, clip some of those whites, and you'll notice the saturation go up too. And then of course you can always punch the saturation also to get even better results.

But I think the RAW images we're getting right now are pretty good. Have you ever seen how "bad" the raw images from the Viper look? They are not pretty either.

please, not to my email, the last time you sent me e-mail, my account got stuck.I got only five megs space.
Could you just send me link to download from?

Exactly, that's what happens with the positive material you see projected.
It has not much more than 256 shades of gray (around 5 stops) and looks great.
The case is worst with positive stock like Kodak Premiere or Fuji High Contrast.Less dynamic range..
If someone saw Moulin Rouge, its copies were made on Premiere.

Hey Ben,

How much was that camera?

Also is the software easy to use to capture material, especially if you're just using RAM?

Sorry Jason I have not seen the Viper stuff...

yes I have seen that 35mm image and it does look bad..but I can see how post can fix that ..bt SMEAR is not somthing you want to fight..its jsut BAD...Jason you know what I am talking about don't you? I seem to remember you pushed an image I sent you with black streaks in the darks?

post that if you can for everyone...that is a bad example because what I shot last night was so much worse that even the raw image had huge streaks in it!!

<<<-- Originally posted by Jason Rodriguez : BTW Obin,

Raw film doesn't look "good" either. If you're going to pack an image with a lot of dynamic range, you're not going to get saturation, and vice versa. Good saturation comes with increased contrast.

Take these images for instance:

http://www.24p.com/asc_web/35mm_girl.jpg

This is a 35mm frame with around 12-13 stops of dynamic range captured. You're not going to get "pretty" images when you do that. Her flesh tones are practically non-existant. Now increase the contrast, clip some of those whites, and you'll notice the saturation go up too. And then of course you can always punch the saturation also to get even better results.

But I think the RAW images we're getting right now are pretty good. Have you ever seen how "bad" the raw images from the Viper look? They are not pretty either. -->>>

All I have to do is point the camera at a window and expose for the inside room and the window will have huge black streaks that go across the dark area of the frame...the camera is at work ior I would shoot it now to show you guys

does anyone know when a camera with the altasense chip is going to be available?

Obin,
A question:
If you are looking at a gamma uncorrected image, how can you know how to expose to the interior light?
Aren't you overexposing?

Just a question...

Wow, this thread has been busy! I'm on a business trip right now, and haven't have Internet access very much, so I apologize to anyone who has sent me e-mail.

Regarding the 10- vs. 12-bit issue. Yes, for some reason the 10-bit data from the SI-1300CL camera comes across padded into a 16-bit word like this:

0000xxxxxxxxxx00

where x = a real data bit.

I don't know why they do it that way, perhaps to make it easy to support 10- and 12-bit cameras with the same code.

Okay,

Here's the "banding" picture

http://home.mindspring.com/~jrod/Banding.jpg

BTW, for this shot this is an extreme example, I guess with Obin it must have been much worse than this, because you would never CC an image to look like this!

Ok, I see now :)

Is it always the same?
What is the supposed way to avoid it?

Yes.

That has been my point with all these images since people started posting them. They're all overexposed! If you're seeing a bright image with a file that has no gamma correction but is suppose to be linear, then you're cramming everything into the top end of the dynamic range of the sensor. Don't be surprised when it doesn't take much to clip!

Has anybody been over to the uncompressed DVX100 page? That is a great example when you see the difference in brightness between the raw 12-bit signal and the 8-bit DV signal (that's been corrected). The 12-bit signal is much darker, so if you're trying to expose the 12-bit like the 8-bit, you're going to be disappointed in the dynamic range that you're getting.

Linear RAW images are not suppose to look good. They're suppose to hold information-just like a film negative.

That is why I'm asking cause I think there is a little problem with the methods people are applying here.
Please try to be very carefull with the way you expose.

The RAW image should be really, really dark, to be able to get good results after correction.

With all the Sony DV cameras, you can see that with a very very bright spot, the CCD gives you a vertical white line so I guess this is a similar problem but CMOS style..

Here's an example of a RAW linear file and the color-corrected version:

http://home.mindspring.com/~jrod/D60_RAW_1.jpg

http://home.mindspring.com/~jrod/D60_RAW_2.jpg

This is from a D60, and the image is blue because I forgot to do the white-balance correctly (you can fix that in Canon's software. BTW, it's also hand-held and I didn't stop down enough to get the edges in focus to keep the shutter high enough so that it wouldn't blur. Top of the Empire State looking towards the Hudson/Chelsea Hotel if anyone is interested). But I just use these two images to illustrate the point of what non-gamm-corrected linear images look like versus the "nice" counterparts we're used to seeing. You can't expose your camera like the second image because it will never look like that as a linear image. The tops of all those buildings would be clipped, and the shadows would look crushed. And you'd complain that it "doesn't look like film" :-)

Good question Juan..this whole thing is hard because the "preview" we get is NOT what is in the raw file..so you must learn how to shoot with this chip...it's not even CLOSE to how you would shoot a DV or DVCPROHD camera..EVERYTHING must be DARK...if you have a hotspot that is close to 100% it will have the smear across the whole image

GREAT example Jason!!!! that is about how I have to shoot the 1300!!!

it's hard to shoot like that because you DON"T know what your really shooting!!!!

Great example Jason!!
As a rule of thumb, when you are dealing with video (high bit depth like we have here) always expose for highlights.

Wow!! that image you posted gives me even great results when I correct it in 8 bit colorspace.I can't believe it.
My gamma was 2.2.

Yes, so the best thing is to either have a histogram showing you what's going on, make your software do the gamma correction for you on it's "preview" window (I'm not sure if the current software does this, Obin this may be something you want to talk to your coder friend about), or use a light meter an an optical viewfinder (paralax involved since we can't see through the lens with these cameras) to judge the scene you're shooting-just like film ;-)

Now Obin,

I don't remember seeing the streaking problems with your first footage that had a lot of hot clipped highlights in them.

Then I believe you switched lenses? And now you're getting streaks?

I'm still wondering if it's the lens and not the chip. CMOS chips are suppose to be much more tolerant of streaking/blow-outs than CCD's because of the way that CMOS doesn't have to do the whole "bucket brigade" thing.

Another thing I'm wondering is Ben can run his camera at 24fps and adjust the shutter from there to reduce or eliminate rolling shutter artifacts. How come you have to run the SI camera at 48fps and drop frames? Maybe if you could also run at 24fps and adjust the shutter from there you wouldn't have that problem.

That's why I'm insisting that without quality lenses anything is posible, and to try to use at least SLR lenses.

BTW,

One more gotcha!

I'm noticing that there's a definite amount of non-linearity in the shadows. Now this may be because I'm not really sure I've gotten a good RAW file before, but still, it seems as though the deepest shadows that come out when you convert your linear file with gamma correction shouldn't actually be showing-So in your conversions, you'd ideally want to adjust the shadow gamma back down, or apply a curves filter that adds a toe to the image. If you skip that step it seems to induce banding in the shadows.

@Juan: That's why I'm insisting that without quality lenses anything is posible, and to try to use at least SLR lenses.

My Angenieux 25mm is definitely a quality lens. I don't know why you have a bug up your a$$ about native c-mount lenses. Should we all shoot with Large Format lenses adapted to our digital SLRs? Sheesh...

You guys, I shot more tests, and I think I'm getting the hang of it. The trick is to turn the gamma all way down to 0.33 (which makes the image brighter -- it's a straight power function. Photoshop takes 1/gamma as the power function. Don't ask.), turn the brightness all the way down, and crank up the contrast until the image gets close to filling the histogram.

I tried not to overexpose any of my new shots, and it wasn't really a problem. The only shot where the sky is blown out was in a really dark little corner of the trees, so I really had to choose between blowing the sky out a little and getting detail in the shadows.

I can't respond to every question at this point, so I'll address a couple:

Juan, it's overcast in Chicago today, and earlier it was much brighter. There was no way to not blow out the sky. Now it's a little darker and I didn't have to make as many compromises.

Re: chroma. That's my new focus. If you mess with the chroma, you shouldn't see too many artifacts. It's when the luma is affected that you can see problems.

Jason: unlike your D60, I can't shoot straight linear with this camera, because it outputs 8bit. You definitely have to do as much as possible before the 10->8 conversion. It does mean a little less flexibility later than if you just dump 12bit from the sensor to your HD.

Someone asked how much the camera is. I told Sumix some people might be coming, and I worked out a slight discount from their normal price. Just mention my name and you'll get it for $1000. It comes with a tripod mount and the USB2 cable, but they said they could also throw in a C-mount adapter for certain mounts -- although they were pretty obscure mounts. ;)

I'll post some new images soon from my latest adventure taking the camera out to the lake. It was rather scary, because it kept threatening to rain. I had plastic for everything just in case, but I had to work fast. I also took stills with my Canon Elph to help in balancing the color, and figuring out how Canon interprets CCD/CMOS data.

- ben

Hey Ben,

Sounds very good. Please post some more stills, cause if this thing works, I really think I'm in. I was going to spend around $2200 on an HDCAM rental for a film I'm working on, but if I can get something like this together for around the same price, I think I'm going with this-after all, I can keep the camera later. Dynamic range and de-bayer stuff is my only consideration right now. Color can be fixed if the information is there-I have pretty good CC software.

How noisy does it get in the shadows? And do you still think it's ISO 100?

BTW, there is a lot you can do with 8-bits, I've done it with HD-CAM many times, and even DVCProHD. If you have a clean 8-bits then you're fine.

Also how easy is it to record to RAM with the software? Is the software as hard to use as X-CAP? Can the software be used with a small touch-screen (1024x768 or even 800x600), or is it hard to use, like lots of little buttons, etc.

Another thing is should a Pentium M at 1.6-1.7Ghz be enough? I guess you'd use a Mini-ITX formfactor.

BTW, you might want to try recording 10-bit linear to 8-bit log-that will keep the shadow information, and clipping highlights won't be so bad. Can you program the 10-bit to 8-bit lookup table by hand?

Ben, the images look pretty good.

I have a couple of tech questions that may allow more usability of that camera:

Can you upload arbitrary luts to the cam to make the 8 bits?
Can I do this once per frame at 24fps ? (SDK, etc ) ( Luts are Tiny )

Thanks
-Les

Jason,

I'd say go for it. I'm starting to get much better images. If it turns out to be a bust, you can always reschedule and shoot with something else. That's the beauty of personal projects. Let me address your other questions:


"Also how easy is it to record to RAM with the software?"

There's a button that says "record to RAM." :) I doesn't get much easier. For the keyboard-inclined, it's Shift-F4. If you want to check it out yourself, you can read the manual or even download the software at this page on Sumix.com.. It might be a bit hard to use at 1024x768 on a touchscreen, but if you talk to Sumix, maybe they'd make a touchscreen version. They've been extremely accomodating, and they're anxious to know what they can do to appeal to filmmakers. I'd say, start a conversation with them.


"Another thing is should a Pentium M at 1.6-1.7Ghz be enough? I guess you'd use a Mini-ITX formfactor."

I'm using a Pentium M at 1.6ghz. It seems to be more than enough. I was wrong about the image adjustments before -- they run in realtime. It's the histogram that sometimes slows things down. I would never use the MiniITX for this application, but whatever. It runs fine on my laptop. I'd much rather have a 14" screen to make focus adjustments than a 7 or 8" touchscreen...


"BTW, you might want to try recording 10-bit linear to 8-bit log-that will keep the shadow information, and clipping highlights won't be so bad. Can you program the 10-bit to 8-bit lookup table by hand?"

I'm going to ask them about this. I'd love to be able to define the LUT myself programmatically.


Les, I don't know why you'd want a new LUT every frame, but maybe. I'm looking into how accessible the LUT is through the SDK. Give me a day or so.

The SDK is on their website -- why not open it up and take a look at the docs? I've gotta convert some files for a while.

- ben

BTW, what's the maximum frame rate at 1280x720? Can the camera even output that frame-size (that's what it looks like you've been sending us), or do you have to crop later on?